Microsoft has committed $2.5 billion to help use AI (and other AI business news).
getty
Getting companies to do something useful with AI is difficult. There’s more news about AI than anyone can read, and companies can do far more with AI than any team has time to test. The hard part is knowing which ones will make a difference for your customers and your staff. If you make the right decision, the tool will pay for itself. If you do it wrong, you could end up wasting months on the wrong thing.
In July, there was some AI news that executives should pay attention to. Two of the world’s biggest companies spent billions of dollars on the same bet within 48 hours. OpenAI has released an agent that can disappear for several hours and come back with all its work completed. The AI industry was ranked for safety for the first time, but the number one companies still fell short. Restaurants can now take orders from within the chat window. And Anthropic has found the secret part of its unique model: solving things before you say a word.
Get the latest AI news and what it means for your business.
July’s AI news that will change what we make and what we sell
Microsoft spent $2.5 billion to help businesses leverage the AI they already have.
On July 2nd, Microsoft Established Frontier Co., Ltd.the $2.5 billion unit is staffed by approximately 6,000 engineers, consultants, and experts who travel to client companies and work alongside their teams to enable them to run AI within their day-to-day operations. Two days earlier, Amazon committed $1 billion to a matching effort within AWS, placing a small group of engineers working within a customer’s business for up to 45 days at a time.
Both companies looked at one bottleneck and came to one conclusion. Too many business owners buy subscriptions and never change the way they run their business. However, companies using AI for their services; worth more than 130%. Your service is the source of growth. Let’s learn AI. Please raise the price.
OpenAI’s new agent runs tasks for hours and returns completed files
OpenAI released on July 9th chat gpt jobsis an agent built on the new GPT-5.6 model that collects information across connected apps and files, breaks down tasks into steps, and continues to run independently for hours to create a finished document, handout, or report. Check in only when a user decision is required. Scheduled tasks allow you to move forward with your projects without anyone having to open the app.
Agents who spend hours working on complete deliverables previously required an enterprise agreement. A free tier is now also available. Choose one report to run each week, give your agent the same summary you would give to a junior employee, and have them run it while you’re on a call that only you can answer.
Restaurants now accepting orders within ChatGPT and Claude
Square rolled out on July 1st new integration This allows you to find and take orders from US food and beverage sellers directly within ChatGPT or a conversation with Claude, without any technical setup required. Orders are sent directly to the merchant’s existing point of sale and kitchen screen. Square does not add marketplace fees to its regular online ordering fees (approximately 2.9% to 3.3% plus 30 cents per transaction). Next up is Alexa.
This is the first clear case of AI chat serving as a sales channel for small and medium-sized businesses. Until this month, I was drafting emails and summarizing documents. Payment is required now. If you sell anything directly to consumers, watch this carefully. The question to answer is whether customers can buy products in the same conversation where they are already asking the AI what to order.
The industry’s most secure AI company earns a C+
On July 7, the Future of Life Institute released its Summer 2026 report. AI Safety Indexscored by seven external judges across 37 scales. Anthropic was ranked #1 among nine major AI companies, receiving only a C+. OpenAI and Google DeepMind each got C.xAI, while DeepSeek and Mistral completely failed. The reviewers found that some companies have softened previous commitments to pause development around set risk thresholds.
Those who ask the world to use tools to perform sensitive tasks admit that their safeguards are behind the curve. Your value is to be the one who checks the output, finds mistakes, and takes responsibility for the results. Use tools, but don’t forget human judgment. Clients pay you to own the results.
Anthropic has found a place for its models to work things out in private
Anthropic magazine published on July 6th new research Identify a small portion of Claude (called J-Space). This part holds the idea of what the model does before it appears in the output. In one test, the team asked Claude to think about the Golden Gate Bridge while copying unrelated sentences and found that “bridge” and “California” were active throughout the passage, even though neither was mentioned in the visible answer. A version trained to interfere with the code displayed the words “fake” and “fraud” before the output appeared normal.
What a model says and what a model does may be different. Now there is a way to see the gap. When you want your AI to suggest what’s important to your client, don’t just rely on the answers on the screen. Read the reason behind it and check out the two juxtapositions. Do not trust the output without checking the contents.
Why Microsoft and Amazon are spending billions to help businesses leverage AI
Companies developing AI have found it difficult to make good use of it, and are pouring large sums of money into the effort. The agent now handles complete jobs on its own. Customers place and pay within the chat window. Models can hide easily overlooked questions beneath confident answers. Deciding what to open is a matter of knowing what to give to the tool and what to keep in your hands.
get mine Free AI playbook For ambitious founders looking to scale.
